Ideas Scripts Chart Stock Screener Forex Signal Finder Economic Calendar How It Works Chart Features House Rules Moderators For the WEB Widgets Stock Charting Library Feature Request FAQ Help & Wiki Twitter

# ANN Strategy v2

10370 56
This is version 2 of my ANN strategy. This version will not repaint, but requires some settings configuration.
The reason why the old version repainted, is because it used the daily OHLC/4 which kept changing, causing lower timeframes ro repaint.
You can now specify which timeframes to use. The bigger timeframe is set in the settings (e.g. 1 day). You set the smaller settings by opening a certain chart (e.g. a 4h chart).
The timeframes are used to calculate the percentual difference, which is fed to the ANN as input

```//@version=2
strategy("ANN Strategy v2")

threshold = input(title="Threshold", type=float, defval=0.0000, step=0.0001)
timeframe = input(title="Timeframe", type=resolution, defval='1D' )

getDiff() =>
yesterday=security(tickerid, timeframe, ohlc4[1])
today=ohlc4
delta=today-yesterday
percentage=delta/yesterday

PineActivationFunctionLinear(v) => v
PineActivationFunctionTanh(v) => (exp(v) - exp(-v))/(exp(v) + exp(-v))

l0_0 = PineActivationFunctionLinear(getDiff())

l1_0 = PineActivationFunctionTanh(l0_0*0.8446488687)
l1_1 = PineActivationFunctionTanh(l0_0*-0.5674069006)
l1_2 = PineActivationFunctionTanh(l0_0*0.8676766445)
l1_3 = PineActivationFunctionTanh(l0_0*0.5200611473)
l1_4 = PineActivationFunctionTanh(l0_0*-0.2215499554)

l2_0 = PineActivationFunctionTanh(l1_0*0.3341657935 + l1_1*-2.0060003664 + l1_2*0.8606354375 + l1_3*0.9184846912 + l1_4*-0.8531172267)
l2_1 = PineActivationFunctionTanh(l1_0*-0.0394076437 + l1_1*-0.4720374911 + l1_2*0.2900968524 + l1_3*1.0653326022 + l1_4*0.3000188806)
l2_2 = PineActivationFunctionTanh(l1_0*-0.559307785 + l1_1*-0.9353655177 + l1_2*1.2133832962 + l1_3*0.1952686024 + l1_4*0.8552068166)
l2_3 = PineActivationFunctionTanh(l1_0*-0.4293220754 + l1_1*0.8484259409 + l1_2*-0.7154087313 + l1_3*0.1102971055 + l1_4*0.2279392724)
l2_4 = PineActivationFunctionTanh(l1_0*0.9111779155 + l1_1*0.2801691115 + l1_2*0.0039982713 + l1_3*-0.5648257117 + l1_4*0.3281705155)
l2_5 = PineActivationFunctionTanh(l1_0*-0.2963954503 + l1_1*0.4046532178 + l1_2*0.2460580977 + l1_3*0.6608675819 + l1_4*-0.8732022547)
l2_6 = PineActivationFunctionTanh(l1_0*0.8810811932 + l1_1*0.6903706878 + l1_2*-0.5953059103 + l1_3*-0.3084040686 + l1_4*-0.4038498853)
l2_7 = PineActivationFunctionTanh(l1_0*-0.5687101164 + l1_1*0.2736758588 + l1_2*-0.2217360382 + l1_3*0.8742950972 + l1_4*0.2997583987)
l2_8 = PineActivationFunctionTanh(l1_0*0.0708459913 + l1_1*0.8221730616 + l1_2*-0.7213265567 + l1_3*-0.3810462836 + l1_4*0.0503867753)
l2_9 = PineActivationFunctionTanh(l1_0*0.4880140595 + l1_1*0.9466627196 + l1_2*1.0163097961 + l1_3*-0.9500386514 + l1_4*-0.6341709382)
l2_10 = PineActivationFunctionTanh(l1_0*1.3402207103 + l1_1*0.0013395288 + l1_2*3.4813009133 + l1_3*-0.8636814677 + l1_4*41.3171047132)
l2_11 = PineActivationFunctionTanh(l1_0*1.2388217292 + l1_1*-0.6520886912 + l1_2*0.3508321737 + l1_3*0.6640560714 + l1_4*1.5936220597)
l2_12 = PineActivationFunctionTanh(l1_0*-0.1800525171 + l1_1*-0.2620989752 + l1_2*0.056675277 + l1_3*-0.5045395315 + l1_4*0.2732553554)
l2_13 = PineActivationFunctionTanh(l1_0*-0.7776331454 + l1_1*0.1895231137 + l1_2*0.5384918862 + l1_3*0.093711904 + l1_4*-0.3725627758)
l2_14 = PineActivationFunctionTanh(l1_0*-0.3181583022 + l1_1*0.2467979854 + l1_2*0.4341718676 + l1_3*-0.7277619935 + l1_4*0.1799381758)
l2_15 = PineActivationFunctionTanh(l1_0*-0.5558227731 + l1_1*0.3666152536 + l1_2*0.1538243225 + l1_3*-0.8915928174 + l1_4*-0.7659355684)
l2_16 = PineActivationFunctionTanh(l1_0*0.6111516061 + l1_1*-0.5459495224 + l1_2*-0.5724238425 + l1_3*-0.8553500765 + l1_4*-0.8696190472)
l2_17 = PineActivationFunctionTanh(l1_0*0.6843667454 + l1_1*0.408652181 + l1_2*-0.8830470112 + l1_3*-0.8602324935 + l1_4*0.1135462621)
l2_18 = PineActivationFunctionTanh(l1_0*-0.1569048216 + l1_1*-1.4643247888 + l1_2*0.5557152813 + l1_3*1.0482791924 + l1_4*1.4523116833)
l2_19 = PineActivationFunctionTanh(l1_0*0.5207514017 + l1_1*-0.2734444192 + l1_2*-0.3328660936 + l1_3*-0.7941515963 + l1_4*-0.3536051491)
l2_20 = PineActivationFunctionTanh(l1_0*-0.4097807954 + l1_1*0.3198619826 + l1_2*0.461681627 + l1_3*-0.1135575498 + l1_4*0.7103339851)
l2_21 = PineActivationFunctionTanh(l1_0*-0.8725014237 + l1_1*-1.0312091401 + l1_2*0.2267643037 + l1_3*-0.6814258121 + l1_4*0.7524828703)
l2_22 = PineActivationFunctionTanh(l1_0*-0.3986855003 + l1_1*0.4962556631 + l1_2*-0.7330224516 + l1_3*0.7355772164 + l1_4*0.3180141739)
l2_23 = PineActivationFunctionTanh(l1_0*-1.083080442 + l1_1*1.8752543187 + l1_2*0.3623326265 + l1_3*-0.348145191 + l1_4*0.1977935038)
l2_24 = PineActivationFunctionTanh(l1_0*-0.0291290625 + l1_1*0.0612906199 + l1_2*0.1219696687 + l1_3*-1.0273685429 + l1_4*0.0872219768)
l2_25 = PineActivationFunctionTanh(l1_0*0.931791094 + l1_1*-0.313753684 + l1_2*-0.3028724837 + l1_3*0.7387076712 + l1_4*0.3806140391)
l2_26 = PineActivationFunctionTanh(l1_0*0.2630619402 + l1_1*-1.9827996702 + l1_2*-0.7741413496 + l1_3*0.1262957444 + l1_4*0.2248777886)
l2_27 = PineActivationFunctionTanh(l1_0*-0.2666322362 + l1_1*-1.124654664 + l1_2*0.7288282621 + l1_3*-0.1384289204 + l1_4*0.2395966188)
l2_28 = PineActivationFunctionTanh(l1_0*0.6611845175 + l1_1*0.0466048937 + l1_2*-0.1980999993 + l1_3*0.8152350927 + l1_4*0.0032723211)
l2_29 = PineActivationFunctionTanh(l1_0*-0.3150344751 + l1_1*0.1391754608 + l1_2*0.5462816249 + l1_3*-0.7952302364 + l1_4*-0.7520712378)
l2_30 = PineActivationFunctionTanh(l1_0*-0.0576916066 + l1_1*0.3678415302 + l1_2*0.6802537378 + l1_3*1.1437036331 + l1_4*-0.8637405666)
l2_31 = PineActivationFunctionTanh(l1_0*0.7016273068 + l1_1*0.3978601709 + l1_2*0.3157049654 + l1_3*-0.2528455662 + l1_4*-0.8614146703)
l2_32 = PineActivationFunctionTanh(l1_0*1.1741126834 + l1_1*-1.4046408959 + l1_2*1.2914477803 + l1_3*0.9904052964 + l1_4*-0.6980155826)

l3_0 = PineActivationFunctionTanh(l2_0*-0.1366382003 + l2_1*0.8161960822 + l2_2*-0.9458773183 + l2_3*0.4692969576 + l2_4*0.0126710629 + l2_5*-0.0403001012 + l2_6*-0.0116244898 + l2_7*-0.4874816289 + l2_8*-0.6392241448 + l2_9*-0.410338398 + l2_10*-0.1181027081 + l2_11*0.1075562037 + l2_12*-0.5948728252 + l2_13*0.5593677345 + l2_14*-0.3642935247 + l2_15*-0.2867603217 + l2_16*0.142250271 + l2_17*-0.0535698019 + l2_18*-0.034007685 + l2_19*-0.3594532426 + l2_20*0.2551095195 + l2_21*0.4214344983 + l2_22*0.8941621336 + l2_23*0.6283377368 + l2_24*-0.7138020667 + l2_25*-0.1426738249 + l2_26*0.172671223 + l2_27*0.0714824385 + l2_28*-0.3268182144 + l2_29*-0.0078989755 + l2_30*-0.2032828145 + l2_31*-0.0260631534 + l2_32*0.4918037012)

buying = l3_0 > 0 ? true : l3_0 < -0 ? false : buying[1]

hline(0, title="base line")
//bgcolor(l3_0 > 0.0014 ? green : l3_0 < -0.0014 ? red : gray, transp=20)
bgcolor(buying ? green : red, transp=20)
plot(l3_0, color=silver, style=area, transp=75)
plot(l3_0, color=aqua, title="prediction")

if (longCondition)
strategy.entry("Long", strategy.long)

if (shortCondition)
strategy.entry("Short", strategy.short)```
Looks fantastic!!!!!!!!!!!!!!!!!!!!. Thanks for sharing!!!! This is really a great effort, appreciate it. Will check tomorrow when market is on.
Thanks for this looking forward to testing!! :)
TYwill test it
Not sure if I'm missing something but when I change the threshold it's not updating?
On smaller time frame of 10 min it gives error "Order limit of 2000 was reached" can you increase this order limit?
sudhir.mehta
i got reply from support team, hope this will help.

Thank you for reaching out TradingView support.
You reported: we are premium user. We have a error during uploading strategy named as Vdub FX Sniper VX3/Strategy and other strategy also has same e...

“Order's limit (2000) was reached” means the number of limits in the strategy has reached the max level of 2000. This limitation is set for a better work of our servers. To plot a strategy starting from a fixed point rather than for the complete data history, please add a comparison time>timestamp(2017, 01, 01, 09, 30) to the order condition, where “time’ is the time of a current bar, “timestamp()” is a feature converting date and time to UNIX format. Please see the function description in Study Script Reference: https://tradingview.com/study-script-reference/#fun_timestamp.

Please see an example of using Vdub FX Sniper VX3/Strategy, in which the strategy will be calculated starting from January 1st 2017. Lines 60-71 are changed:

//============ signal Generator ==================================//
Piriod=input('720')
ch1 = security(tickerid, Piriod, open)
ch2 = security(tickerid, Piriod, close)
longCondition = crossover(security(tickerid, Piriod, close),security(tickerid, Piriod, open))
if (longCondition and time>timestamp(2017, 01, 01, 09, 30))
shortCondition = crossunder(security(tickerid, Piriod, close),security(tickerid, Piriod, open))
if (shortCondition and time>timestamp(2017, 01, 01, 09, 30))
strategy.entry("SELL", strategy.short)

///////////////////////////////////////////////////////////////////////////////////////////

Best regards,

470 Olde Worthington Road, Suite 200, Westerville, OH 43082 USA
I was using the first version with SPX500, EURUSD, USOIL ... I was finding higher profitability rates on Version 1 than I am on 2 .... why is that? V2 Gives me under 40% profitability while version 1 gives me over 80% .... doesn't make sense to use V2 in this case, right?
pcmourao
@pcmourao, because V1 is repainted its better backtesting while V2 is better for real time trading as repainting paints a rosy picture and is not accurate
just tested it in realtime.
it does repaint!!!
i think you should work on ohlc, only ohlc cause repaint.
metmet
"today=ohlc4"
metmet
today=ohlc4(a bar ago, im not able to write square brackets)
metmet
Only open would not repaint, but that would be misusing the network as it was not trained to do this.

I tested this real quick and netto profit went down from 2653\$ to 1426\$
sirolf2009
i use it with 15' and 1' calculated on the last COMPLETE bar.
i love this code.
thank you a lot.
(point=square bracket)

yesterday=security(tickerid, timeframe, ohlc4.1.)
today=ohlc4.1.
metmet
in this way no repaint at all
metmet
You may want to change the first ohlc4 to ohlc4.2.
What you're doing now is calculating the difference between yesterday and yesterday. Which is and always will be 0, rendering the indicator useless
sirolf2009
oh no wait, nevermind I somehow forgot how to read code :)
This is interesting though, it took the netto profit from 2653\$ to 4553\$
sirolf2009
it's the difference between 15' and 1'
:)
metmet
I've changed the code but what timeframe is your chart and what inputs do you have in the setting?
metmet
Can you repost the code with the new addition that does not repaint? Thanks a lot
metmet
What do you mean 15' and 1' Where? 15 minute and 1 minute timeframe?
sirolf2009
Sorry sirolf2009, just a question: Could you make a strategy from an existent indicator in here? Would you be able to add in it a buy/sell/stop/profit condition?
sirolf2009
@sirolf2009, Can you post the new code? That would be awesome! :)
metmet
Could you post the script that does not repaint?
metmet
@metmet, OHLC gathers better data for future predictions, I've been working on this indicator for almost 2 years and I finally built the biggest ANN ever. ANN meaning the meaning of that. It's so much data it takes 5-7 minutes to load on each chart.
Hello Sirolf2009. Great coding on version 1 of this strategy and the repaint doesn't really bother as this doesn't seem to be a strategy for day trading but works really well on the 4H timeframe, which is more my timeframe for trades. I do have one thing I wanted to ask and that is what is the base for the strategy.... I mean how does it calculate the BUY and SELL signals, what conditions have to be met? I guess the question is if this strategy wasn't coded, what are the conditions that have to be met to generate the signals....almost like if I was looking for the signals "manually" instead of what he script does.
pcmourao
This is a question i also want to learn ?
fxcandle
This is not an easy question, if you look at the source, you'll see a ton of random number in the center. These numbers are all pulled through some functions, then they're added up together and then they're pulled through some functions again. If you were to calculate it by hand, you'll probably spend around 10 minutes calculating the final value. From there it's easy, if the final value is more than 0, the indicator is bullish, if the value is less than 0, the indicator is bearish. If it switches between bullish and bearish, a signal is put out
With the Heinkin Ashi candles this script give better profits. More stable I think.
Coinhike
HA is problematic on tradingview, the backtest uses the wrong data
Coinhike
Any time you use a Heikin Ashi chart with a strategy, the results you get are not correct. It's using the HA open price as the trade entry/exit price, and those prices are not the "actual" price at the time of the trade. Sorry to rain on your parade lol
Hi Sirolf, buying variable should use the l3_0 value of the last bar with : l3_0. Otherwise it is repainted.
Coinhike
Hey man, I know but that's not how the indicator should be used. You're actually supposed to wait until the end of the day to make your trades, you can trade halfway through the day, but you'll risk the repainting (which is like a breakeven point)
sirolf2009
Hey sirolf, what do you mean by the end of the day? because the market is open 24/5 all week weekdays, and we're all on different timezones. Thanks in advance!
Hi Sirolf, thank you for this wonderful strategy! Just a question: if I would like to use it with a 15min chart, should I just open a 15min chart and apply the v2 strategy, or should i apply some more settings? Thanks a lot. Alessandro
Cool strategy, unfortunately any time you use a longer resolution than your current tick, TradingView gives you the OHLC values from the END of that interval. You must specify OPEN price (because that will be correct since it's from the start of the interval) but all the other values will be using "future" data. That's why your signals repaint. Want to see my 100% accurate multi-timeframe strategy? It's only 1 line of code. Lol.
I would warn anyone who uses this or any other strategy that uses a custom resolution to STOP USING IT or make sure you specify OPEN price. THE RESULTS ARE NOT REAL PEOPLE..............
yesterday would be ohlc running with daily candles on 60 minute ticks, today would be ohlc OHLC values don't get locked in until the end of the daily period, which is 24 hours long.

zmm20
Ok that's weird, the brackets got deleted from my message. It should say - ohlc brackets 48 and ohlc brackets 24. Why don't brackets come through in messages?
zmm20
https://imgur.com/r4jVnzn here is my multi-timeframe indicator. Too bad it's not real! Yours is doing the same thing.
zmm20
I'm fully aware of this and currently working on different methods to fix it. I've spoken with a few people who are trading with this indicator, and they handle the repainting as a breakeven point. I've tried using the open price as input, but the ANN simply was not trained to be used this way and I got very poor results out of it.
sirolf2009
what's your final thoughts? should we wait for an update?
Vincent121
ANN still seems profitable if you trade every repaint, which is a notion that has been confirmed by multiple people. Currently I'm looking at different ways to use ANN which may or may not result in new publications in the future
sirolf2009
Okay, can you please keep up updated?
sirolf2009
what if we wait till the very end of the current candle formation(but before it actually ends) and act only then.
Isn't it pretty much the way it's done during backtesting? And this should help to eliminate repainting problem
Hi, does this use the OHLC4 close of the time-frame the user is currently on?

To me it looks like this line "today=ohlc4" means that if the user is on a 5min chart that it compares the ohlc4 of the daily with the current ohlc4 value of whatever time-frame the trader is on.

Therefore if the user is on a hourly chart this might work better than on a minute chart.
ChartArt
I wanted to say: I understand this code that it does the following:

If the user is on a 5min chart that it compares the ohlc4 of the daily from one day ago with the current ohlc4 value of the 5min chart. Therefore comparing a daily bar with a 5min bar.
ChartArt
Yeah, it uses the timeframe you're currently on. But it was trained using daily data which is why I chose the somewhat weird names for my variables
Hey man try your strategy out with Renko charts. Granted it was taking results from 1995, but AMD gave a profit factor of 43! AKS gave a profit factor of 29 too! That's the way to use this, not with regular candles.
thomas.gigure
We checked this out on both Renko and Heikin Ashi. However, the trades tradingview makes are based on the Renko closes, not the real closes. Meaning that the trades it makes are impossible to replicate in the "real" world. We checked how it performed when it calculates using Renko and performed trades on regular candles and it's actually worse than performing calculations on regular candles
Thanks for the updates. how to turn back the "Buy" and "sell" signals as it shows in v1?
Hello,

How do I get this to send me an alert when it changes from green to red? thank you
Hmm is it even fixed yet =/
Equity curve caught my eye. Unfortunately the data it uses is 'unknowable' and results could not be reproduced in real time since one does not know the current OHLC. Given that it is looking forward on higher time frame the impossible results should be better than they are. Somewhat interesting code but unfortunately fantasy result.
Hello, is it possible to get a sound alert when a trade is triggered. If so how?
launell
@launell,
I have the same question
I doubt this works now with version 3 of pine script, not even bothering to test it, but I assume it's a dud with V3.